the cryosphere discuss., doi:10.5194/tc-2016-72, …...103 and methodology, presentation of results...

1

The increasing snow cover in the Amur River Basin from MODIS observations during 2000-2014 1

2

*1Xianwei Wang,

1Yuan Zhu,

2Yaning Chen,

1Hailing Zheng,

3Henan Liu,

1Huabing Huang,

1Kai Liu 3

and 1,4

Lin Liu 4

5

1Center of Integrated Geographic Information Analysis, and Guangdong Key Laboratory for 6

Urbanization and Geo-simulation, School of Geography and Planning, 7

Sun Yat-sen University, Guangzhou, China, 510275 8

2State Key Laboratory of Desert and Oasis Ecology, Xinjiang Institute of Ecology and Geography, 9

Chinese Academy of Sciences, Urumqi, China, 830011 10

3Heilongjiang Climate Data Center, Ha’erbin, China, 150001 11

4Department of Geography, University of Cincinnati, Cincinnati, OH, USA 12

13

14

Corresponding Authors: Xianwei Wang 15

Email: [email protected]; [email protected] 16

Tel: 86-20-84114623 17

18

The Cryosphere Discuss., doi:10.5194/tc-2016-72, 2016Manuscript under review for journal The CryospherePublished: 12 May 2016c© Author(s) 2016. CC-BY 3.0 License.

2

Abstract 19

The study applies the improved cloud-free Moderate Resolution Imaging Spectral radiometer 20

(MODIS) daily snow cover product (MODMYD_MC) to investigate the snow cover variations 21

from snow Hydrologic Year (HY) HY2000 to HY2013 in the Amur River Basin (ARB), northeast 22

Asia. The fractions of forest cover were 38%, 63% and 47% in 2009 in China (the southern ARB), 23

Russia (the northern ARB) and ARB, respectively. Forest demonstrates complex influences on the 24

snow accumulation and melting processes and on optical satellite snow cover mapping. Validation 25

results show that MODMYD_MC has a snow agreement of 88% against in situ snow depth (SD) 26

observations (SD≥4cm). The agreement is about 10% lower at the forested stations than at the 27

non-forested stations. Snow Cover Durations (SCD) from MODMYD_MC are 20 days shorter than 28

ground observations (SD≥1cm) at the forested stations, while they are just 8 days shorter than 29

ground observations (SD≥1cm) at the non-forested stations. Annual mean SCDs in the forested 30

areas are 21 days shorter than those in the nearby farmland in the SanJiang Plain. SCD and Snow 31

Cover Fraction (SCF) are negatively correlated with air temperature in ARB, especially in the snow 32

melting season, when the mean air temperature in March and April can explain 86% and 74% of the 33

mean SCF variations in China and Russia, respectively. From 1961 to 2015, the annual mean air 34

temperature presented an increase trend by 0.33℃/decade in both China and Russia, while it had a 35

decrease trend during the study period from HY2000 to HY2013. The decrease of air temperature 36

resulted in an increase of snow cover although the increase of snow cover was not significant above 37

the 90% confidence level. SCD and SCF had larger increase rates in China than in Russia, and they 38

were larger in the forested area than in the nearby farmland in similarly climatic settings in the 39

SanJiang Plain. The air temperature began to increase again in 2014 and 2015, and the increasing 40

air temperature in ARB projects a decrease of snow cover extent and periods in the coming years. 41

Keywords: MODIS; Snow cover; Increasing; Amur River Basin; 42


3

1. Introduction 43

Seasonal snow covers about 45% of the Northern Hemisphere land areas in winter, e.g., 46% 44

and 47% in January of 2014 and 2015, respectively (Robinson, 2015). Mean monthly snow-cover 45

extent in the Northern Hemisphere has declined at a rate of 1.3% per decade over the last 40 years, 46

and most decrease occurred in spring and early summer (Barry et al., 2009). In March and April, the 47

mean snow cover extent in Northern Hemisphere decreased 1.6%±0.8% per decade in the period 48

from 1967 to 2012 (IPCC, 2013). Snow cover variations in the spring season have critical 49

consequences for surface energy and water budgets and significant implications for cold region 50

ecosystems (Yang et al., 2003; Chapin et al., 2005; Brown et al., 2007). The projected continuing 51

increase of air temperature also suggests more snow cover reduction for mid latitudes by the end of 52

this century (IPCC, 2013). Reduction of snow cover extent acts as a positive feedback to global 53

warming by reducing the land surface albedo, and also affects the structure of ecosystem due to the 54

interactions of plants and animals with snow (Barry et al., 2009). 55

The Amur River forms a large part of the border between Russia and China, and near half of the 56

Amur River Basin (ARB) is within the Heilongjiang, Inner Mongolia and Jilin Provinces of China. 57

This region is one of the three primary snow distribution centers in China. About one fifth of water 58

resources in ARB are attributed to snowfall, and most areas are covered by snow over four months 59

each year (Chen et al., 2014; Wang et al., 2014). According to the land cover product of 60

MCD12Q1 in 2009 (Friedl et al., 2010), forest was about 47% in ARB and has a complex influence 61

on snowpack duration (Varhola et al., 2010; Dickerson-Lange et al., 2015; Moeser et al., 2016). In 62

the past half century, more and more wetland and forest had been drained and converted to farmland 63

in the Sanjiang Plain between Ussuri River and Songhua River (main tributaries of the Amur River) 64


4

in the Heilongjiang Province, China (Wang et al., 2006; Xu et al., 2012). The cultivated land area 65

increased from 7, 900 km² to 52, 100 km², and population density increased from 13 to 78 persons 66

per km² from 1949 and 2000; in contrast, the wetlands decreased from 53, 400 km² to 9, 070 km² 67

during the same period (Li et al., 2004; Wang et al., 2009a). The increase in agricultural land and 68

population brought about severe pressure on the local ecosystem, such as climate and water 69

regulation, support of habitats, and food provisions (Dale and Polasky, 2007; Wang et al., 2015). 70

Several studies investigated the snow cover variations in Northeast China based on the satellite 71

snow cover products (Zhang, 2010; Lei et al., 2011; Chen et al., 2014; Wang et al., 2014), but few 72

studies examined the snow cover variations covering the entire ARB or compared the snow cover 73

variations in the forested and non-forested areas. Lei et al. (2011) evaluated the snow classification 74

accuracy of the Moderate Resolution Imaging Spectral radiometer (MODIS) 8-day standard snow 75

cover products and the Advanced Microwave Scanning Radiometer-EOS (AMSR-E) snow water 76

equivalent (SWE) in the Heilongjiang Province during 2002-2007. Zhang (2010) directly used the 77

daily MODIS Terra snow cover product to compute the snow-covered days in Northeast China by 78

only counting the snow-covered pixels under clear sky, and omitted the snow-covered pixels 79

blocked by cloud. Chen et al. (2014) analyzed the spatiotemporal variations of snow in Northeast 80

China by using the multiday combinations of MODIS snow cover products, which have a mean 81

cloud cover of 4% and mean combination period of 7 days, similar to the MODIS standard 8-day 82

maximum combination product. Cloud is one of the major factors influencing the MODIS snow 83

cover mapping, leading to underestimate the snow-covered areas. Several methods have been 84

developed to reduce or remove the cloud cover in the MODIS snow cover products (Parajka and 85

Bloschl, 2008; Liang et al., 2008b; Xie et al., 2009; Hall et al., 2010; Parajka et al., 2010; Paudel 86


5

and Andersen, 2011; Wang et al., 2014). Based on previous approaches, Wang et al. (2014) 87

developed an algorithm to generate a daily cloud-free snow cover product from MODIS Terra and 88

Aqua standard daily snow cover products and tested it in the Central Heilongjiang Province, China. 89

This cloud-free snow cover product provides better representation of timely snow cover variations 90

in the tested areas than previous studies in this region (Zhang, 2010; Lei et al., 2011; Chen et al., 91

2014; Wang et al., 2014). The daily cloud-free snow cover product and the accordingly derived 92

snow maps provide critical information for snow disaster prevention and mitigation, agriculture and 93

water resources management, and are key inputs for hydrologic and climatic models as well (Wang 94

et al., 2014). 95

Therefore, the primary objectives of this study are to 1) validate the MODIS daily cloud-free 96

snow cover product developed by Wang et al (2014) using more snow depth measurements in the 97

entire Heilongjiang Province, China; and 2) apply this cloud-free snow cover product to study the 98

spatiotemporal variations of the snow cover extent in the entire Amur River Basin, especially 99

comparing the snow cover variations between the Northern ARB (Russian part) and the southern 100

ARB (China Part), and evaluating the influence of forest on snow cover mapping and variations. 101

The rest of this paper is followed by a brief description of the study area, statement on primary data 102

and methodology, presentation of results, summary and discussions. 103

104

2. Study area 105

The Amur River (Called HeiLongJiang in China) Basin (ARB) is located in Northeast Asia 106

(42°～57° N, 108°～141° E) crossing China, Russia and Mongolia, with a total area of 2.08×106 107

km² (Fig.1). The southern portion of ARB is located in China (43% of the ARB area), the northern 108


6

and eastern part is in Russia (49%), and only a small part in the southwestern upstream is in 109

Mongolia (9%). The Amur River originates in the mountains of Great Hinggans in China and the 110

Cherskogo in Russia, firstly flows towards northeast, then slowly turns to southeast in a great arc 111

after the upstream Shilka River and Argun River meet together, and finally changes back towards 112

northeast and pours into the Strait of Tartary, the Sea of Okhtsk. The Songhua River and Ussuri 113

River in the south are the two primary tributaries of Amur River in China, forming the Song-Liao 114

Plain and the Sanjiang Plain, respectively. Elevations in most areas are lower than 500 m and 115

partially rise up to 2500 m in the northern and eastern mountain peaks. Forests were 47% in ARB, 116

38% in China and 63% in Russia in 2009, and mainly distributes in the northern ARB and in 117

mountains. Grassland is mainly in Mongolia and China (Fig. 1). In the past half century, a large part 118

of wetlands had been converted into farmland in the Sanjiang Plain in China, which provides great 119

amount of wheat, corn, rise, green bean, etc. (Li et al., 2004). Two areas (around 3500km2

each) of 120

farmland and forest in the Sanjiang Plain are arbitrarily selected to illustrate the snow cover 121

variations (Polygons A and B in Fig. 1). 122

The climate of ARB is controlled by the high latitude East Asian monsoon, the Siberian cold, 123

dry winds from west in winter and the warm, humid ocean winds from east in summer. The annul 124

mean air temperature and precipitation in the upstream, middle stream and downstream are around 125

-2℃, 1℃, and 0℃, and 400mm, 600mm and 650mm, respectively (Yu et al., 2013). The freezing 126

days with mean air temperature below 0 ℃ is about 150 days per year in most areas, and it is up to 127

210 days in the northern mountains and in the Great Hinggan Mountains. Observed snow depth 128

were over 1 m in some regions. Snow covered durations in most stations in the Heilongjiang 129

Province in China ranges from 80-120 days per year, and even longer than 160 day at several 130

stations (Yu et al., 2013). 131


7

3. Data and Methodology 132

3.1 Snow depth and air temperature 133

Meteorological data are collected at 83 national standard weather stations in Heilongjiang 134

Province, China, including snow depth, precipitation and air temperature from 2000 to 2010 (Fig. 1). 135

Snow depth data are daily observations rounded to 1 cm and used to validate the MODIS snow 136

cover products. Four stations are screened out due to data quality and inconsistency, and total 79 137

stations are used for validation. In general comparison, one MODIS pixel value collocated with one 138

weather station is extracted to compare with the corresponding in situ snow depth data. To minimize 139

the impacts of the spatial representative error of in situ measurements, the snow depth data ≥ 4 140

cm are used to validate the accuracy of MODIS snow cover mapping, and other data of 0 cm (no 141

snow) and 1-3cm (patchy snow) are also provided for assistance (Parajka and Bloschl, 2008; Wang 142

et al., 2008, 2009b). 143

In a detailed investigation within snow hydrologic years (HY) from September 1, 2007 to 144

August 31, 2008 (HY2007) with minimum snow and HY2009 with maximum snow, the averages of 145

different-size pixel patches are compared with the snow depth data centered at each one of the four 146

weather stations, representing cultivated farmland (S31), natural grassland (S53) and forests (S12 147

and S52). The pixel patches consist of pixels 1×1, 3×3, 5×5, 9×9, and 21×21 pixels, 148

corresponding to areas of 0.5 km×0.5 km, 1.5 km×1.5 km, 2.5 km×2.5 km, 4.5 km×4.5 km, and 149

10.5 km×10.5 km. 150

Besides the snow depth data at the 83 stations in Heilongjiang Province, daily mean air 151

temperature data at 18 stations in China and 22 stations in Russia are downloaded from the Global 152

Historical Climatology Network (GHCN), http://www.ncdc.noaa.gov/oa/climate/ghcn-daily. Some 153


8

of the GHCN stations are same to part of the 83 meteorological stations (Fig. 1). The time spans 154

from 1961 to 2015 for daily air temperature. 155

3.2 MODIS standard snow cover products 156

Identical MODIS instruments are carried on both Terra and Aqua satellites, which were 157

launched in December 1999 and May 2002 and pass the equator at the local time of 10:30 am and 158

1:30 pm, respectively. A series of standard snow cover products are generated using the same 159

algorithm from both Terra and Aqua MODIS reflectance data (Hall et al., 2002, 2010; Wang et al., 160

2009b). These standard products include the daily MOD10A1 (Terra) and MYD10A1 (Aqua) and 161

the 8-day maximum composite of MOD10A2 and MYD10A2 with 500m spatial resolution and 10 162

degree by 10 degree tiles, and the daily, 8-day maximum and monthly mean snow cover products of 163

MOD10C1-3 and MYD10C1-3 with 0.05° global climate model grid (Wang et al., 2014). 164

The 500-m MODIS Terra/Aqua daily MOD/MYD10A1 snow cover products from 2000 to 165

2014 are used in this study. Six MODIS tiles are required to cover the entire the Amur River Basin. 166

These tiles are h24v03, h25v03, h25v04, h26v03, h26v04 and h27v04. All MODIS data are 167

downloaded from the National Snow and Ice Data Center (NSIDC) (http://nsidc.org/data/modis). 168

Snow cover variations are analyzed mainly in a snow hydrological year, beginning on September 1 169

and ending on next August 31. For instance, HY2013 refers to a period from September 1 of 2013 170

to August 31 of 2014 (Wang and Xie, 2009). 171

3.3 Improved MODIS snow cover product 172

Cloud blockage is a severe impediment for optical satellite snow cover mapping. For examples, 173

the annual mean cloud fraction is about 50% and can be up to 95% in several continuous days in the 174


9

Northern Xinjiang, China and on the Colorado Plateau, U.S. (Wang et al., 2008; Xie et al., 2009). 175

Fortunately, snow cover remains relatively stable in a couple of hours and even days especially in 176

winter compared to the cloud cover. The three-hour difference that satellites Terra and 177

Aqua/MODIS pass the equator allows us to combine the morning and afternoon observations of 178

MODIS Terra/Aqua to obtain a cloud-less snow cover product (Parajka and Bloschl, 2008; Wang et 179

al., 2009b; Xie et al., 2009). The improved cloud-free MODIS daily snow cover product 180

(MODMYD_MC) is generated from the MODIS Terra/Aqua standard daily snow cover product by 181

a cloud-removal algorithm (Wang et al., 2014). This algorithm included two automated steps. Firstly, 182

the daily snow cover images of MOD10A1 and MYD10A1 are combined to generate a daily 183

maximum snow cover image called MODMYD_DC. Then, the cloud pixels on the current-day 184

MODMYD_DC (n) are replaced by the previous-day MODMYD_DC (n-1) cloud-free pixels on 185

day (n-k) until all cloud pixels are replaced, finally generating a daily cloud-free snow cover 186

product. The number of cloud-persistent days (k) k is normally less than 5 days in the tested tile of 187

h26v04 although it is not limited in the replacement (Wang et al., 2014). 188

3.4 MODIS land cover product 189

MODIS land cover product of MCD12Q1 in 2009 is used to classify the land cover type in 190

ARB (Friedl et al., 2010). MCD12Q1 was generated by ensemble supervised approaches with 191

various classification systems. MCD21Q1 have five land cover classifications in each calendar year. 192

The 17-class International Geosphere-Biosphere Programme (IGBP) classification is applied in this 193

study. The land cover types at each station include deciduous coniferous forest, deciduous forest, 194

shrub, grass and crops, and are reclassified into forest (coniferous forest, deciduous forest and shrub) 195

and non-forest (grass and crops). In the entire ARB, all land cover types are also classified into 196


10

forest and non-forest to compare the snow cover variations in the forested and non-forested areas. 197

198

4. Results 199

4.1 Assessment of snow cover classification 200

The in situ snow depth measurements at 79 meteorological stations are used to validate the 201

snow classification for the four MODIS snow cover products, including two daily standard products 202

(MOD10A1 and MYD10A1), one daily combination (MODMYD_DC) and one multi-day 203

combination (MODMYD_MC). Snow depth measurements are divided into three groups: snow 204

depth ≥ 4cm (snow) , 1cm ≤ snow depth ≤ 3cm (fractional snow) and snow depth = 0cm (snow-free 205

land). Fig. 2 shows the mean agreements of the four products against in situ snow depth (≥4cm) 206

measurements during the 10 snow hydrologic years (HY2000-HY2009). The large proportion 207

(~50%) of cloud blocks the MODIS instrument to retrieve the land cover under cloud, resulting low 208

agreements of snow (34-38%) and land (64-66%) mapping for the daily Terra and Aqua MODIS 209

snow cover products. The daily combination (MODMYD_DC) of MOD10A1 and MYD10A1 210

reduces cloud by 10% and increases snow and land agreements by 5-10%. For the improved daily 211

cloud-free snow cover product MODMYD_MC, agreements of snow and land classification are 88% 212

and 92%, respectively. The land and overall agreements from all 79 stations are similar to and the 213

snow agreement is slightly lower than those from the 53 stations under the MODIS tile of h26v04 in 214

the Central Heilongjiang Province, China (Wang et al., 2014). 215

The lower value of snow agreement is likely related to the fact that there are more stations 216

within forests. Fig. 3a shows that the snow agreements at forested stations is about 10% lower than 217

those at non-forested stations for the four snow cover products although their mean snow depth at 218


11

the forested stations is 5 cm larger, which is because most forested stations are located in higher 219

latitude and altitudes. For the fractional snow cover (1cm ≤ snow depth < 4cm), the agreements for 220

MOD10A1, MYD10A1, MODMYD_DC and MODMYD_MC are 20%, 12%, 23% and 59% at 221

non-forested stations, respectively (Fig.3b). Compared to the two daily standard products, the snow 222

agreement of MODMYD_MC increases by 40% for the fractional snow. Similarly, the snow 223

agreements are lower at the forested stations than at the non-forested stations. 224

Fig. 4 illustrates mean agreements of snow cover duration (SCD) in each snow hydrologic year 225

from HY2000 to HY2009 between MODMYD_MC and in situ observations at 79 individual 226

stations. The agreement is computed by equation (1). 227

A = (1 −|SCDmod−SCDgrd|

SCDgrd) × 100% (1) 228

where A represents agreement, SCDgrd is snow-covered days (SCD) from ground measurements 229

with snow depth > 1cm, SCDmod is SCD recorded by MODIS (Wang et al., 2009b, 2014). 230

Generally, SCD from MODMYD_MC agrees well (91±6%) with and is 8±8 days shorter 231

than ground observations at the non-forested stations. This is similar to other regions, such as 232

Northern Xinjiang (Wang et al., 2009b), Colorado Plateau (Xie et al., 2009), Pacific Northwest 233

(Gao et al., 2010). In contrast, the mean SCD agreement is 79±13% and 20±23 days shorter than 234

ground observations at the forested stations. 235

4.2 Representative errors 236

The above analysis shows large agreement variations of snow mapping at individual stations, 237

especially within different land covers. In order to investigate these disagreements at different 238

stations and assess the representative of a single station measurements or a single pixel value within 239


12

different land covers, four typical stations (Fig.1) in the same sub-region are selected for detailed 240

analysis, representing cultivated farmland (S31), natural grassland (S53) and forest (S12 and S52). 241

Figure 5 displays the daily in situ snow depth measurements and MODMYD_MC retrievals at the 242

selected four stations from HY2000 to HY2009. Table 1 summarizes the ground measured snow 243

depth and the according MODMYD_MC retrievals. 244

At the flat stations of farmland and grassland, MODIS (MODMYD_MC) retrievals are in good 245

agreement with the in situ snow depth observations (Fig.5a, b, Table 1), e.g., the Snow Cover Onset 246

Date (SCOD) and Snow Cover End Date (SCED) for the continuous winter snow-covered period 247

were consistent with ground observations, except for some transient snowfall events before SCOD 248

and after SCED. Those missed transient snowfall events were caused primarily by the persistent 249

cloud blockage that snow melts away before it turns clear (Wang et al., 2009b). This indicates that 250

winter snow unanimously covers those areas and the point measurements at the clearing 251

meteorological sites have good spatial representative. At different pixel patches from 1×1, 3×3, 5252

×5, 9×9, to 21×21 pixels (0.25, 2.25, 6.25, 20.25 to 110.25km2), the mean values of MODIS 253

retrievals are consistent and agree well with the ground snow depth measurements especially in the 254

snow-rich year of HY2009 (Tables 2 & 3). The higher agreements in the snow-rich year of HY2009 255

again demonstrate the importance of spatial representative of snow depth measurements in 256

validating the satellite retrievals. 257

The forested stations of S12 and S52 represent a good case and a poor case of agreements, 258

respectively. At station S12, the maximum snow depth in the snow-least (HY2007) and snow-rich 259

years (HY2009) were 14 and 39 cm, respectively. Accordingly, the agreement in HY2009 was 260

higher than that in HY2007 for the different pixel patches, especially for the fractional snow (Table 261


13

4). Meanwhile, the agreements did not decrease evidently with the increase of pixel patches except 262

for on December 31, 2007. Fig. 5c also shows that MODIS performed poorly for HY2001, HY2002 263

and HY2007, when there were less snow and snow depth were less than 10 cm mostly. In contrast, 264

the snow depth in each year at S52 was much less than those at S12 (Fig. 5d). MODIS failed to 265

retrieve the snow in many periods when the ground observed snow but had snow depth less than 4 266

cm (Fig. 5d). In the two years of HY2007 and HY2009, MODIS agreements even increased with 267

the increase of pixels (Table 5). This again shows that part of the disagreement is caused by the poor 268

representative of ground measurements at the point scale within forested areas. It is quite 269

challenging to accurately map the snow cover and snow depth in the forested areas 270

(Dickerson-Lange et al., 2015). For the forested areas, a larger sampling area may better represent 271

the overall snow cover conditions. 272

4.3 Snow cover variations 273

4.3.1 Snow cover distributions 274

The improved cloud-free MODIS daily snow cover product (MODMYD_MC) is used to study 275

the spatiotemporal variations of snow cover in the entire Amur River Basin. Fig. 6 presents the 276

spatial distribution of the mean snow-covered days during the 14 snow hydrologic years from 277

HY2000 to HY2013 in ARB. SCDs are classified into seven categories: shorter than 30 days, 30 to 278

60 days, 61 to 90 days, 91 to 120 days, 121 to 150 days, 151 to 180 days and longer than 180 days. 279

Overall, latitude (a proxy of available solar radiation) dominates the SCD distribution. The higher 280

the latitude, the longer the mean SCD is. For example, SCD is longer than 180 days in Northern 281

ARB but less than 60 days in the southern central ARB. Elevation is the secondary factor 282

determining SCD. Mountainous areas have longer SCD than the nearby plain. SCDs in mountains 283


14

of Stanovoi, Tukuringer and Bureinskiy Mountain are normally longer than 150 days, while they 284

are less than 120 days in the neighboring plains. 285

Meanwhile, the forested areas show shorter SCD than the non-forested areas with similar 286

latitude and elevation, such as in the Great Hinggans and their neighboring plains (Fig. 1 and Fig. 6). 287

This shorter SCD within forest could be explained from two sides. On one hand, the snow cover 288

extent is likely underestimated by the MODIS optical remote sensing techniques, and this 289

underestimate ranges from 9% to 37% across different forest densities (Raleigh et al., 2013). On the 290

other hand, there is actually shorter SCD within forest than in the nearby farmland due to the 291

complex influence of the forest on snow accumulation, redistribution, and ablation processes 292

(Varhola et al., 2010). 293

The histogram of the mean SCD in ARB shows a near normal distribution (Fig. 7). The 14-year 294

mean SCDs range from 0 to 260 days in ARB. The primary part is between 90 and 150 days, only 295

10% is less than 60 days, and 8% is longer than 180 days. The basin-scale mean SCD is 121 days in 296

ARB. Most areas are snow covered from December to March, and part of them is from October or 297

November to April (Fig. 8). Snow normally starts to cover the basin from the north and in 298

mountains in October, and then melts away in April and early May. 299

4.3.2 Snow cover variations 300

Mean SCF in ARB has large inter-annual variations, and the largest SCF variation ranges 301

(max-min) are in the snow accumulation period of November (29%) and December (42%) and in 302

the snow melting months of March (53%) and April (39%) (Fig. 8). All maximum snow cover 303

extent occurred in the snow accumulation and melting months of HY2012 (Fall of 2012 and Spring 304

of 2013), while the minimum snow in the accumulation and melting periods were in HY2001 (Fall 305


15

of 2001) and HY2007 (Spring of 2008), respectively. 306

The detailed variations of snow cover in ARB are illustrated in three snow hydrologic years by 307

four plots, SCD anomaly (a), monthly mean SCF (b), SCOD in fall (c), and SCED in spring (d) in 308

Figs. 9. These three snow hydrologic years represent a snow-poor year in HY2001, a snow-neutral 309

year in HY2005, and a snow-rich year in HY2012 (Fig. 9-11). SCD anomalies in each year are 310

grouped into seven categories with a 30-day interval: <-60 days, -60 to -31 days, -30 to -11 days, 311

-10 to 10 days (supposed to be no change), 11- 30 days, 31-60 days, and >60 days. Red colors 312

express shorter SCD below average, and the white indicate longer SCD above average (e.g., Fig. 313

9a). SCOD and SCED are also grouped into seven categories with a half month interval, from 314

October 15th

to January 1st for SCOD, and from February 15

th to May 1

st for SCED. Similar to Fig. 315

9a, red colors represent later SCOD and earlier SCED, i.e., shorter SCD, and the white represent 316

earlier SCOD and later SCED, leading to longer SCD (e.g., Fig. 9c & d). 317

For a snow-poor year in HY2001, SCD was below the 14-year mean in most ARB (red colors) 318

except for some small areas, e.g., in the upstream of Zeya catchment in Northern ARB (Fig. 9a). 319

Monthly mean SCF was much lower than the 14-year mean in November, December, February and 320

March (Fig. 9b). In the Northern ARB, snow began to cover the land late October 2001 and began 321

to melt in early April 2002. In contrast, snow did not begin to cover the land until late December 322

2001 and melted away in February 2002 in the Southern and Western ARB (Fig. 9c & 9d). 323

For a snow-neutral year in HY2005, SCD was around the 14-year mean (dark color) in most 324

ARB. SCD was about 15 days longer than the mean in the west, central north and the downstream 325

basin, and it was 30 days shorter than the mean in southern ARB, and even 60 days shorter than the 326

mean in the Southwestern part in Mongolia (Fig. 10a). SCF was same to the 14-year mean in each 327


16

month (Fig. 10b). Snow normally began to cover the northern, central and southern ARB from 328

October, November and late December, and then melted away from southern ARB to Northern ARB 329

from early February, March and late April (Fig. 10c & 10d). 330

For a snow-rich year in HY2012, SCD was above the 14-year mean in most ARB, especially in 331

the southern plains (white color) and was on the opposite of HY2001 (Figs. 9a & 11a). SCF was 332

much larger than the 14-year mean from December 2012 to April 2013 (Fig. 11b). Snow began to 333

cover the entire ARB in October and November, and only some forested areas did not cover by 334

snow until early December, e.g., in the Great Xingans (Fig. 11c). Meanwhile, snow did not begin 335

melt away until late March and April in most areas, and until early May in the northern ARB (Fig. 336

11d). 337

In summary, snow cover shows large spatial and inter-annual variations. The plains and 338

cultivated farmland show much larger inter-annual variations than the mountains and forests. The 339

combination plots of SCD, SCF, SCOD and SCED together illustrate the detailed snow cover 340

variations and are significant in snow water resource, agriculture and disaster risk management. 341

4.3.3 Snow cover changes 342

During the 14 snow hydrological years from HY2000 to HY2013, annual SCD and the 343

March-April mean snow cover fraction (SCF) had a similar increasing trend in ARB, Russia and 344

China (Fig. 12, Table 6). The monthly mean SCF in ARB has larger variations in March and April 345

than in other months (Fig. 8). HY2012 had the largest SCF/SCD, while HY2001 and HY2007 had 346

the smallest SCF/SCD (Fig. 12). The mean SCF in March and April of 2008 showed the smallest 347

snow cover during the 14 years. Both SCD and SCFs had large inter-annual fluctuations and a 348

slightly increasing trend. China had a larger increasing rate (slope) than Russia, and March had a 349


17

larger increase rate than other months in China, although they were not significant above the 90% 350

confidence level in all areas (Table 6). 351

SCF was controlled dominantly by air temperature in the melting season and negatively 352

correlated with air temperature, especially in March and April, when r2

were 0.86 in China and 0.74 353

in Russia in the 14 years (Fig.13). The increasing SCF from HY2000 to HY2013 is further 354

supported by the decreasing air temperature recorded from meteorological stations in China and 355

Russia in the same period, although the air temperature had an increase trend from 1961 to 2015 356

(Fig. 14). The mean air temperature reached a high record in 2007 and then continuously decreased 357

till 2013, resulting in an increasing SCF especially in HY2012. It began to increase again in 2014 358

and 2015. The increasing air temperature projects a further decrease of snow cover extent and 359

periods. 360

Besides the administrative boundary between Russia and China, the snow cover variations are 361

further investigated in the forested and non-forested areas. There were shorter (mean = -12 days) 362

SCD in the forested areas than in the non-forested areas although most forested areas are located in 363

the northern part and have higher elevations than the non-forested areas in the entire ARB (Fig. 15a). 364

Snow in the forested areas starts slightly later (-4 days) and melts away earlier (8 days) than those in 365

the non-forested area. During the 14 years, SCD had a slightly increasing trend in both forested and 366

non-forested areas in the entire ARB (Fig. 15a, Table 7). The increasing SCD occurred in earlier 367

SCOD and later SCED. The change rates in the forested areas were smaller than those in the 368

non-forested areas, and only SCOD changes (advance) in the non-forested areas were significant at 369

the 90% confidence level (Table 7). 370

Moreover, there were much shorter (-21 days) SCD in the forested areas than in the 371


18

non-forested areas in the Sanjiang Plain (Fig. 15b). Both areas (around 3 500km2) are located in a 372

same latitude ranges and selected to compare the snow cover variations. The forested area is about 373

200-300m higher than the neighboring farmland area. During the 14 years, SCD had a slightly 374

increasing trend in both forested and non-forested areas in the Sanjiang Plain (Fig. 15b, Table 8). 375

The snow starts later (14 days) and melts earlier (-6 days) in the forested area than in the 376

non-forested area. The change rates were larger in the forested areas than in the non-forested areas, 377

although these changes are not significant above the 90% confidence level but close to the 378

significant r thresholds, especially in the forested areas (Table 8). 379

380

5. Summary and Discussions 381

5.1 Effects of forest on snow 382

5.1.1 Representative errors of ground measurements 383

The improved daily cloud-free MODIS snow cover shows high agreements with the ground 384

snow depth measurements (Fig.2), while the snow agreements within the 79 stations is slightly 385

lower than those at the 53 stations (Wang et al., 2014). This is because most of the excess stations 386

are located within forest. The snow agreements are about 10% lower within the forested stations 387

than at the non-forested stations (Fig.3). MODIS SCDs within the forest stations are shorter than the 388

ground observations (Fig.4b), and also have lower agreements and r values with ground 389

observations than those at the non-forested stations (Fig.4a). 390

This suggests that forest is the dominant contributions to the disagreement. The point 391

observations of snow depth at the clearing meteorological sites may not represent the areal 392


19

distributions of snow accumulations in the neighboring forested areas (500m for a MODIS pixel). 393

The forests shows different snow accumulation and melt dynamics compared to the clearcuts. There 394

is likely less snow accumulation in the forested ground since a large part of snowfall is intercepted 395

by the forest canopy/branches (Varhola et al., 2010). For instance, peak snow accumulations 396

increased significantly from the forest edge to the nearby clearcuts of 0.5, 1 and 2 times of the forest 397

height (Golding and Swanson, 1986). Accurate estimation of snow accumulation in forested 398

watersheds is problematic, and a single snow depth measurement cannot validate a MODIS pixel 399

with confidence, especially in forested pixels (Raleigh et al., 2013). The network of dense ground 400

sensors of temperature and snow depth observations can provide good spatial representative of 401

snow accumulations under forest canopy only at limited campaign sites (e.g., Raleigh et al., 2013; 402

Dickerson-Lange et al., 2015), while the snow depth measurements at single stations are the merely 403

available data in the ARB watershed and in other large watersheds world around as well. In the 404

common point-pixel comparison, the mean value of a larger pixel patch, instead of the 405

station-located one pixel, could be a better representative to MODIS retrievals especially within the 406

forested area in some cases (Tables 4 & 5). 407

5.1.2 Impacts of forest on MODIS snow cover mapping 408

Besides the ground representative errors at a single clearing forest station, MODIS snow cover 409

mapping algorithm encounters challenges in the forested areas (Xin et al., 2012). The improved 410

algorithm of MODIS binary snow cover mapping in the forest mask uses the Normalized 411

Differences of Snow Index (NDSI) and the minimum reflectance (0.1) of MODIS band 4 (545-565 412

nm, green band) (Hall et al., 1995; Klein et al., 1998). Snow-covered forests usually have a much 413

lower reflectance in the green band than pure snow. Both NDSI and the reflectance of band 4 can be 414


20

reduced by the existence of forest stands due to the shadow and tree-snow mixture effects, resulting 415

in less snow cover estimate (Musselman et al., 2012; Dickerson-Lange et al., 2015). This 416

underestimate can be explained from two aspects. Firstly, MODIS sensors have a ±55°wide 417

across-track scan angle, which makes the view zenith angle approach 65°at the end of the scan 418

line. The large view zenith angles reduce the capability of MODIS sensor to see the forest gaps (the 419

snow-covered ground) through the tree canopy and trunks, and these gaps are essentially detectable 420

at lower view zenith angle or near the nadir view (Liu et al., 2008). 421

Secondly, in the mid and high latitude regions like ARB, the solar elevation angles are low in 422

most of snow-covered winter months. Forest cast larger shadows and shaded areas with lower solar 423

elevation angles, resulting in lower reflectance and NDSI. Our results (not shown) demonstrates 424

that MODIS snow agreements in the forested stations in Heilongjiang Province in China are 12% 425

lower in December and January (with lower elevation angles) than in February and March (with 426

higher elevation angles), while the snow agreements are similar during both periods at the 427

non-forested stations. Snow normally persists through this period from December to March at most 428

stations. As a consequence, MODIS snow cover mapping accuracy is lower in the forested areas 429

than in the non-forested areas (Hall et al., 2002; Xin et al., 2012; Raleigh et al., 2013). This again 430

illustrates the challenges in accurately mapping snow cover extent in the forested areas using 431

MODIS and other optical remote sensors. MODIS-retrieved snow extent is the best available snow 432

cover product in the forested areas at the regional and global scales with relative long-term and 433

consistent observations. The improved MODIS daily cloud-free snow cover product could present 434

the overall snow cover extent within forested areas although there is a large uncertainty. 435

5.1.3 Effects of forest on snow accumulation and ablation 436


21

SCD was shorter in the forested areas than in the nearby non-forested areas (Fig. 6). In the 437

entire ARB, SCD was 12 days shorter in the forested area than in the non-forested areas although 438

most of forested areas are distributed in the northern ARB and in higher elevation zones (Fig. 15a). 439

At the similar climatic settings in the Sanjiang Plain, SCD was 21 days shorter in the forested areas 440

than in the non-forested areas (Fig. 15b). Part of this shorter SCD in the forested areas could be 441

explained by the less estimate of MODIS snow cover mapping algorithm (Xin et al., 2012; Raleigh 442

et al., 2013). 443

The dominant cause is likely due to the less snow accumulation within the forested areas than 444

on the nearby farmland, e.g., in the two case areas of forest and farmland in the Sanjiang Plain (Figs. 445

15b). Depending on specific conditions, up to 60% of cumulative snowfall could be intercepted and 446

sublimated back to atmosphere (Pomeroy, 1998). Snow accumulated in forested areas was up to 40% 447

lower than that in nearby clearing sites (e.g. Winkler et al., 2005; Jost et al., 2007). Meanwhile, the 448

snowmelt rates could be up to 70% lower in forests than in open areas due to the shelter and 449

shading effect of forest on ground snow (e.g. Varhola et al., 2010). The lower snowmelt rate would 450

cause snow persisting about 10 days longer under the forest canopy than in the nearby clearings 451

although there might be less snow accumulation under forest canopy, e.g. in the study sites of Sierra 452

Nevada (Raleigh et al., 2013). There are complex interactions between forest and snow. Whether 453

SCD persists longer or shorter than the nearby clearings is found to be a function of latitude, aspect, 454

forest structure, climate and seasonal meteorology (Molotch et al., 2009; Musselman et al., 2012). 455

5.2 Snow cover variations 456

The MODIS snow cover extent showed a slightly increase trend during the 14 snow 457

hydrological years from HY2000 to HY2014 in ARB, primarily consisting of Russia and China (Fig. 458


22

12). The northern ARB (Russia) had a smaller increase rate than the southern ARB (China), and 459

March and April had a larger increase rate than other months in China (Table 6). Since the forest 460

cover fractions in China, Russia and ARB were 38%, 63% and 47%, respectively, this leads to a 461

larger increase rate in the non-forested area than in the forested area in the entire ARB (Table 7). In 462

contrast, in the similarly climatic conditions at the two nearby areas in the SanJiang Plain (Fig. 1), 463

the forested areas showed much shorter SCD and slightly larger increase rates than in the 464

non-forested area (Table 8). 465

The increase trends of snow cover in ARB is consistent with the study of Chen et al. (2014) that 466

the snow cover extent and precipitation in the northeast China (mostly in the southern ARB) 467

showed an increase trend during the 10 years from HY2004 to HY2013, while air temperature 468

presented a decrease trend. During the snowy season, snow cover extent was negatively correlate 469

with the air temperature (Chen et al., 2014), especially in the melting season of March and April, 470

when the variations of air temperature can explain 74% and 86% of the mean SCF in Russia and 471

China, respectively (Fig. 13). Compare to the reference period from 1961 to 1990, the annual mean 472

air temperature in the periods of 1991-2000 and 2001-2015 increased by 1.0℃ and 1.1℃ in China 473

and 0.9℃ and 1.1℃ in Russia, respectively. The air temperature showed a decrease trend since 474

2007 and did not begin to increase until 2014 (Fig. 14), resulting in an increasing SCF from 475

HY2000 to HY2013, especially in HY2012 (Figs. 12 and 15). 476

The change trends of snow cover and air temperature in ARB during the recent 15 years in the 477

21st

century are different from the long-term trend from 1950 to 2010, when annual mean air 478

temperature increased significantly by 1.4-2.4℃ in different regions, and precipitation only had an 479

increase trend in the ARB downstream region (Yu et al., 2013). According to the in situ snow depth 480


23

measurements, Li et al. (2009) found that SCOD retreated by 1.9 days per decade and SCED 481

advanced 1.6 days per decade from 1961 to 2005 in the Heilongjiang Province, China, although the 482

cumulative snow depth showed an increase trend (Chen and Li, 2011). The changes of snow cover 483

duration occurred primarily in the plains, while increases of snow depth took place in mountains (Li 484

et al., 2009). The snow changes in ARB were also different from the monthly mean snow cover 485

extent in the North Hemisphere, which decreased about 5% from 1960s to 2000s (Barry et al., 486

2009). In March and April, the mean snow cover extent in Northern Hemisphere decreased 487

1.6%±0.8% per decade in the period from 1967 to 2012 (IPCC, 2013). Although the snow cover 488

extent in ARB demonstrated a different change pattern during the recent 14 years compared to the 489

long-term trend since 1950s in the context of global warming, the increasing air temperature in 490

ARB projects a decrease of snow cover extent and periods in the coming years. 491

492

6. Conclusions 493

The study applies the improved cloud-free MODIS daily snow cover product (MODMYD_MC) 494

to investigate the snow cover variations from HY2000 to HY2013 in the Amur River Basin. The 495

fractions of forest cover were 38%, 63% and 47% in 2009 in China (the southern ARB), Russia (the 496

northern ARB) and ARB, respectively. Forest demonstrates complex impacts on the snow 497

accumulation and melting processes and on optical satellite snow cover mapping. Validation results 498

show that MODMYD_MC has a snow agreement of 88% against in situ snow depth observations 499

(≥4cm). The agreement is about 10% lower at the forested stations than at the non-forested stations 500

although the former mean snow depth is 5 cm larger. SCDs from MODMYD_MC are 20 days 501

shorter than in situ observations at the forested stations, while they are closer (-8 days) to those at 502


24

the non-forested stations. Annual mean SCDs from MODMYD_MC in the forested areas were 503

greatly shorter (-21days) than those in the nearby farmland in the SanJiang Plain. 504

SCD/SCF is negatively correlated with air temperature in ARB, especially in the snow melting 505

season, when the mean air temperature in March and April can explain 86% and 74% of the mean 506

SCF variations in China and Russia, respectively. From 1961 to 2015, the annual mean air 507

temperature presented an increase trend by 0.33℃/decade in both China and Russia, while it had a 508

decrease trend during the study period from HY2000 to HY2013. The decrease of air temperature 509

resulted in an increase of snow cover although the increase of snow cover was not significant above 510

the 90% confidence level. SCD/SCF had larger increase rates in China than in Russia, and they 511

were larger in the forested area than in the nearby farmland in the similarly climatic settings in the 512

SanJiang Plain. The air temperature began to increase again in 2014 and 2015, and the increasing 513

air temperature in ARB projects a decrease of snow cover extent and periods in the coming years. 514

Acknowledgements: 515

This study is funded by the Open Foundation of the State Key Laboratory of Desert and Oasis 516

Ecology, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences 517

(#G2014-02-06), Natural Science Foundation of China (#41371404), Fundamental Research Funds 518

from Sun Yat-sen University (#15lgzd06), National Basic Research Program of China 519

(#2012CB955903) and the Startup Foundation for Oversea Returnees from the Department of 520

Education of China (#2013693). We thank the US NASA and NSIDC for providing the MODIS 521

land cover and snow cover products and the Heilongjiang Climate Data Center for the in situ snow 522

depth data. 523

524


25

References: 525

Barry, R.G., Armstrong, R.., Callaghan, T., Cherry, J., Nolin, A., and Russell, D. Chapter 4: Snow. Global 526

Outlook for Ice and Snow, United Nations Environment Programme. ISBN: 978-92-807-2799-9, Job No: 527

DEW/0924/NA, 2009. 528

Brown, R., Derksen, C., Wang, L.: Assessment of spring snow cover duration variability over northern Canada 529

from satellite datasets, Remote Sens. Environ., 71, 297–308, 2007. 530

Chapin, F., Sturm, M., Serreze, M., McFadden, J., Key, J., Lloyd, A., et al.: Role of land-surface 531

changes in Arctic summer warming, Science, 310, 657−660m, 2005. 532

Chen, G. and Li, D.: Temporal-spatial characteristics of cumulative snow depth in Northeast China 533

and its vicinity, Meteorological Monthly, 37(5), 513-521, 2011 (In Chinese). 534

Chen, S., Yang, Q., Xie, H., Zhang, H., Lu, P. and Zhou, C.: Spatiotemporal variations of snow 535

cover in northeast China based on flexible multiday combinations of moderate resolution 536

imaging spectroradiometer snow cover products, Journal of Applied Remote Sensing, Vol.8, 537

084685-15,2014.. 538

Dale, V.H. and Polasky, S.: Measures of the effects of agricultural practices on ecosystem services, 539

Ecol. Econ., 64, 286–296, 2007. 540

Dickerson-Lange, S. E., Lutz, J.A., Martin, K.A., Raleigh, M.S., Gersonde, R. and Lundquist, J.D.: 541

Evaluating observational methods to quantify snow duration under diverse forest canopies, 542

Water Resour. Res., 51, doi: 10.1002/2014WR015744, 2015. 543

Friedl, M.A, Sulla-Menashe, D., Tan, B., Schneider, A., Ramankutty, N., Sibley, A. and Huang, X.: 544

MODIS Collection 5 global land cover: Algorithm refinements and characterization of new 545

datasets, Remote Sensing of Environment, 114, 168–182, 2010. 546


26

Gao, Y., Xie, H., Yao, T. and Xue, C.: Integrated assessment on multi-temporal and multi-sensor 547

combinations for reducing cloud obscuration of MODIS snow cover products of the Pacific 548

Northwest USA, Remote Sensing of Environment, 114, 1662-1675, 2010. 549

Golding, D. and Swanson, R.: Snow distribution patterns in clearings and adjacent forest, Water 550

Resour. Res., 22 (13), 1931–1940, 1986. 551

Hall, D.K., Riggs, G.A. and Salomonson, V.V.: Development of methods for mapping global snow 552

cover using moderate resolution imaging spectroradiometer data, Remote Sensing of 553

Environment, 54, 127–140, 1995. 554

Hall, D.K., Riggs, G.A., Salomonson, V.V., DiGirolamo, N.E., Bayr, K.J. 2002. MODIS 555

snow-cover products. Remote sensing of Environment, 83, 181-194. 556

Hall, D.K., Riggs, G.A., Foster, J.L. and Kumar, S.V.: Development and evaluation of a 557

cloud-gap-filled MODIS daily snow-cover product, Remote Sensing of Environment, 114, 558

496-503, 2010.. 559

IPCC: Summary for Policymakers. In: Climate Change 2013: The Physical Science Basis. 560

Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental 561

Panel on Climate Change [Stocker, T.F., D. Qin, G.-K. Plattner, M. Tignor, S. K. Allen, J. 562

Boschung, A. Nauels, Y. Xia, V. Bex and P.M. Midgley (eds.)]. Cambridge University Press, 563

Cambridge, United Kingdom and New York, NY, USA, 2013. 564

Jost, G., Weiler, M., Gluns, D. and Alila, Y.: The influence of forest and topography on snow 565

accumulation and melt at the watershed-scale, J. Hydrol., 347, 101-115, 2007. 566

Klein, A.G., Hall, D.K. and Riggs, G.A.: Improving snow cover mapping in forests through the use 567

of a canopy reflectance model, Hydrological Processes, 12, 1723–1744, 1998. 568

Lei, X., Song, K., Wang, Z., Zhong, G., Zhao, K., Liu, D. and Zhang, B.: Accuracy evaluation of 569


27

MODIS and AMSR-E snow cover products in the Heilongjiang drainage basin, Journal of the 570

Graduate School of the Chinese Academy of Sciences, 28, 43-50, 2011 (In Chinese). 571

Liang, T., Zhang, X., Xie, H., Wu, C., Feng, Q., Huang, X. and Chen, Q.: Toward improved daily 572

snow cover mapping with advanced combination of MODIS and AMSR-E measurements, 573

Remote Sensing of Environment, 112, 3750-3761, 2008. 574

Li, F., Zhang, B. and Zhang, S.: Ecosystem service valuation of Sanjiang Plain, Journal of Arid 575

Land Resources and Environment, 18(5), 19-24, 2004 (In Chinese). 576

Li, D., Liu, Y., Yu, H. and Li, Y.: Spatial-Temporal Variation of the Snow Cover in Heilongjiang 577

Province in 1951 -2006, Journal of Glaciology and Geocryology, 31 (6), 1011-1018, 2009. (In 578

Chinese) 579

Liu, J., Melloh, R.A., Woodcock, C.E., Davis, R.E., Painter, T.H. and McKenzie, C.: Modeling the 580

view angle dependence of gap fractions in forest canopies: implications for mapping fractional 581

snow cover using optical remote sensing, Journal of Hydrometeorology, 9, 1005–1019, 2008. 582

Moeser, D., Mazzotti, G., Helbig, N. and Jonas, T.: Representing spatial variability of forest snow: 583

Implementation of a new interception model, Water Resour. Res., 52, doi: 584

10.1002/2015WR017961, 2016. 585

Molotch, N.P., Brooks, P.D., Burns, S.P., Litvak, M., Monson, R.K., McConnell, J.R. and 586

Musselman, K.N.: Ecohydrological controls on snowmelt partitioning in mixed-conifer 587

sub-alpine forests, Ecohydrology, 2 (2), 129–142, 2009. 588

Musselman, K.N., Molotch, N.P., Margulis, S.A., Kirchner, P.B. and Bales, R.C.: Influence of 589

canopy structure and direct beam solar irradiance on snowmelt rates in a mixed conifer forest, 590

Agricultural and Forest Meteorology, 161, 46 – 56. http: //dx.doi.org 591

/10.1016/j.agrformet.2012.03.011, 2012. 592


28

Ott, R. L., Larson, R., Rexroat, C. and Mendenhall, W.: Statistics: a Tool for the Social Sciences, 593

PWS-Kent Publishing Company, Boston, MA, USA,1992. 594

Parajka, J. and Bloschl, G.: Spatio-temporal combination of MODIS images–potential for snow 595

cover mapping, Water Resources Research, 44, W3406, 2008. 596

Parajka, J., Pepe, M., Rampini, A., Rossi, S. and Blöschl, G.: A regional snow-line method for 597

estimating snow cover from MODIS during cloud cover, Journal of Hydrology, 381, 203-212, 598

2010. 599

Paudel, K.P. and Andersen, P.: Monitoring snow cover variability in an agropastoral area in the 600

Trans Himalayan region of Nepal using MODIS data with improved cloud removal 601

methodology, Remote Sensing of Environment, 115, 1234-1246, 2011. 602

Pomeroy, J.W., Parviainen, J., Hedstrom, N. and Gray, D.M.: Coupled modelling of forest snow 603

interception and sublimation, Hydrol. Proc., 12, 2317–2337, 1998. 604

Raleigh, M.S., Rittger, K., Moore, C.E., Henn, B., Lutz, J.A. and Lundquist, J.D.: Ground-based 605

testing of MODIS fractional snow cover in subalpine meadows and forests of the Sierra Nevada, 606

Remote Sens. Environ., 128, 44–57, 2013. 607

Robinson, D.: Northern Hemisphere continental snow cover extent: 2014 update. Rutgers 608

University, Retrieved on September 15, 2015 at website: http://climate.rutgers.edu /snowcover, 609

2015. 610

Varhola, A., Coops, N.C., Weiler, M. and Moore, R.D.: Forest canopy effects on snow 611

accumulation and ablation: An integrative review of empirical results, J. Hydrol., 392(3-4), 612

219–233, 2010. 613

Wang, Z., Zhang, B., Zhang, S., Li, X., Liu, D., Song, K., Li, J., Li, F. and Duan, H.: Changes of 614

land use and of ecosystem service values in Sanjiang Plain, Northeast China, Environ Monit 615


29

Assess, 112, 69–91, 2006. 616

Wang, X., Xie, H. and Liang, T.: Evaluation of MODIS snow cover and cloud mask and its 617

application in Northern Xinjiang, China, Remote Sensing of Environment, 112, 1497-1513, 618

2008. 619

Wang, J., Lu, X., Jiang, M., Li, X. and Tian, J.: Synthetic evaluation of wetland soil quality 620

degradation: A case study on the Sanjiang Plain, Northeast China, Pedosphere, 19:756–64, 621

2009a. 622

Wang, X., Xie, H., Liang, T. and Huang, X.: Comparison and validation of MODIS standard and 623

new combination of Terra and Aqua snow cover products in northern Xinjiang, China, 624

Hydrological Processes, 23, 419-429, 2009b. 625

Wang, X. and Xie, H.: New methods for studying the spatiotemporal variation of snow cover based 626

on combination products of MODIS Terra and Aqua. Journal of hydrology, 371, 192-200, 627

2009. 628

Wang, X., Zheng, H., Chen, Y., Liu, H., Liu, L., Huang, H. and Liu, K.: Mapping snow cover 629

variations using a MODIS daily cloud-free snow cover product in Northeast China, Journal of 630

Applied Remote Sensing, 8(1), 084681. doi:10.1117/1.JRS.8.084681, 2014. 631

Wang, Z., Mao, D., Li, L., Jia, M., Dong, Z., Miao, Z., Ren, C. and Song, C.: Quantifying changes 632

in multiple ecosystem services during 1992-2012 in the Sanjiang Plain of China, Science of the 633

Total Environment, 514, 119-130, 2015. 634

Winkler, R.D., Spittlehouse, D.L. and Golding, D.L.: Measured differences in snow accumulation 635

and melt among clearcut, juvenile, and mature forests in southern British Columbia, Hydrol. 636

Proc., 19, 51–62, 2005. 637

Xie, H, Wang, X. and Liang, T.: Development and assessment of combined Terra and Aqua snow 638


30

cover products in Colorado Plateau, USA and northern Xinjiang, China, Journal of Applied 639

Remote Sensing, 3, 1-14, 2009. 640

Xin, Q., Woodcock, C.E., Liu, J., Tan, B., Melloh, R.A. and Davis, R.E.: View angle effects on 641

MODIS snow mapping in forests, Remote Sensing of Environment, 118, 50-59, 2012. 642

Xu, Y.M., Li, Y., Ouyanga, W., Hao, F.H., Ding, Z.L. and Wang, D.L.: The impact of long-term 643

agricultural development on the wetlands landscape pattern in Sanjiang Plain, Procedia 644

Environmental Sciences, 13 (2012), 1922-1932, 2012. 645

Yang, D., Robinson, D., Zhao, Y., Estilow, T. and Ye, H.: Streamflow response to seasonal snow 646

cover extent changes in large Siberian watersheds, Journal of Geophysical Research, 108(D18), 647

4578−4591, 2003. 648

Yu, L.L., Xia, Z.Q., Li, J.K. and Cai, T.: Climate change characteristics of Amur River, Water 649

Science and Engineering, 6(2): 131-144, 2013. 650

Zhang, H.: Study on Spatio-temporal Variations of Snow from 2000 to 2009 in Northeast China. 651

Master Thesis. Jilin University, 2010. (In Chinese) 652

653


31

Table 1. Summary of snow and snow-free land recorded by ground measurements and MODMYD_MC retrievals 654

for ten years from 2000 to 2010 at four typical stations representing cultivated farmland, natural grassland and 655

forests. The number in the parenthesis is percentage. 656

Stations Land

cover/use

Elevation

(m)

Latitude

(degree)

Longitude

(degree)

Days

(SD≥1cm)

Snow from

MODIS

Days

(SD≥4cm)

Snow from

MODIS

Suiling, S31 Farmland 202 47.14 127.06 1136 1058 (93%) 857 839 (98%)

Jiamusi, S53 Grassland 81 46.49 130.17 1213 1128 (93%) 1026 991 (97%)

Yichun, S12 Forest 240 47.44 128.55 1302 806 (88%) 1093 751 (92%)

Yilan, S52 Forest 100 46.15 129.350 1047 597 (57%) 644 444 (69%)

657

658

Table 2. The snow cover fraction in different pixel patches centered at stations S31 (cultivated farmland) in snow 659

hydrologic years of 2007/9/1-2008/8/31 (HY2007) with least snow and 2009/9/1-2010/8/31 (HY2009) with most 660

snow. 661

Snow depth

(cm)

Hydrologic

Year

Ground

SCD

Snow cover fraction from MODMYD_MC at S31

1×1 3×3 5×5 9×9 21×21

SD=0 HY2007 280 1% 1% 1% 1% 1%

HY2009 217 1% 1% 1% 1% 1%

1≤SD≤3 HY2007 36 72% 77%/ 75% 73% 74%

HY2009 11 82% 76% 77% 77% 79%

SD≥4 HY2007 49 97% 98% 97% 97% 98%

HY2009 137 100% 100% 100% 100% 100%

Max Snow Depth = 13cm on Dec 31, 2007 100% 100% 100% 95% 97%

Max Snow Depth = 26cm on Mar 16, 2010 100% 100% 100% 100% 100%

Forest Fraction from MCD12Q1 in 2009 0% 0% 0% 0% 0%

662

+663


32

Table 3. The snow cover fraction in different pixel patches centered at stations S53 (grassland) in snow hydrologic 664

years of 2007/9/1-2008/8/31 (HY2007) with least snow and 2009/9/1-2010/8/31 (HY2009) with most snow. 665

Snow depth

(cm)

Hydrologic

Year

Ground

SCD


1×1 3×3 5×5 9×9 21×21

SD=0 HY2007 296 3% 1% 1% 2% 3%

HY2009 218 3% 2% 2% 3% 3%

1≤SD≤3 HY2007 4 25% 18% 45% 14% 12%

HY2009 5 60% 45% 19% 39% 41%

SD≥4 HY2007 65 94% 96% 92% 88% 89%

HY2009 142 95% 95% 95% 91% 92%




666

667

Table 4. The snow cover fraction in different pixel patches centered at stations S12 (forest) in snow hydrologic 668

years of 2007/9/1-2008/8/31 (HY2007) with least snow and 2009/9/1-2010/8/31 (HY2009) with most snow 669

Snow depth

(cm)

Hydrologic

Year

Ground

SCD


1×1 3×3 5×5 9×9 21×21

SD=0 HY2007 256 3% 2% 1% 2% 4%

HY2009 202 3% 1% 3% 2% 3%

1≤SD≤3 HY2007 39 51% 63% 58% 58% 55%

HY2009 8 88% 85% 87% 86% 86%

SD≥4 HY2007 55 87% 86% 82% 80% 78%

HY2009 140 87% 88% 87% 87% 85%




670

671

672

673


33

Table 5. The snow cover fraction in different pixel patches centered at stations S52 (forest) in snow hydrologic 674

years of 2007/9/1-2008/8/31 (HY2007) with least snow and 2009/9/1-2010/8/31 (HY2009) with most snow. 675

Snow depth

(cm)

Hydrologic

Year

Ground

SCD


1×1 3×3 5×5 9×9 21×21

SD = 0 HY2007 292 1% 1% 2% 2% 2%

HY2009 207 0% 2% 2% 2% 2%

1≤SD≤3 HY2007 29 0% 18% 24% 31% 46%

HY2009 18 0% 14% 22% 26% 39%

SD ≥ 4 HY2007 44 2% 34% 40% 57% 75%

HY2009 140 59% 62% 64% 76% 88%




676

677

Table 6. The linear trend/slope (s), interception (b) and correlation coefficients (r) of the mean Snow Cover 678

Fraction (SCF) with time (year) from HY2000 to HY2013 in the Amur River Basin (ARB), Russia and China for 679

Fig 12b. The r threshold for statistical significance is 0.45 at the 90% confidence levels for n=14 (Ott et al., 1992). 680

r s (percent/year) b (Percent)

ARB Russia China ARB Russia China ARB Russia China

Annual 0.27 0.25 0.34 0.30% 0.26% 0.46% 29.60% 35.30% 22.30%

Mar-Apr 0.18 0.09 0.32 0.56% 0.28% 1.10% 42.30% 59.10% 25.50%

Mar 0.22 0.11 0.36 0.78% 0.30% 1.67% 61.10% 77.70% 41.30%

681

682


34

Table 7. The linear trend/slope (s), interception (b) and correlation coefficients (r) of snow cover duration (SCD), 683

Snow Cover Onset Dates (SCOD) and Snow Cover End Dates (SCED) with time (year) from HY2000 to HY2013 684

in the forested and non-forested areas in ARB for Fig 15a. The r threshold for statistical significance is 0.45 at the 685

90% confidence levels for n=14 (Ott et al., 1992). The negative r (smaller) values for SCOD mean that snow 686

cover starts earlier. 687

Forested Areas Non Forested Areas

SCD SCOD SCED SCD SCOD SCED

Mean (day) 115 329 79 127 325 87

r 0.23 -0.23 0.20 0.35 -0.50 0.19

s (day/year) 0.97 -0.47 0.50 1.31 -0.75 0.43

b (Julian day) 108 332 75 118 331 83

688

Table 8. The linear trend/slope (s), interception (b) and correlation coefficients (r) of snow cover duration (SCD), 689

Snow Cover Onset Dates (SCOD) and Snow Cover End Dates (SCED) with time (year) from HY2000 to HY2013 690

in the forested and non-forested areas in the Sanjiang Plain, China for Fig 15b. The r threshold for statistical 691

significance is 0.45 at the 90% confidence levels for n=14 (Ott et al., 1992). The negative r (smaller) values for 692

SCOD mean that snow cover starts earlier. 693

Forested Areas Non Forested Areas

SCD SCOD SCED SCD SCOD SCED

Mean (day) 99 344 78 120 330 84

r 0.39 -0.35 0.33 0.26 -0.13 0.31

s (day/year) 1.61 -0.70 0.91 1.36 -0.47 0.90

b (Julian day) 86 334 70 109 350 77

694

695

696


35

697

Figure 1. The Amur River Basin in Northeast Asia, including 79 meteorological stations in the Heilongjiang 698

Province of China and 40 GHCN stations, the Amur River, tributaries and ranges. 699

700


36

701

Figure 2. The mean agreement of four snow cover products (MOD10A1, MYD10A1, 702

MODMYD_DC, MODMYD_MC) and their cloud fraction (the cloud fraction of MODMYD_MC 703

= 0). On the horizontal X axis, “Snow” and “Land” refer to the agreement (%) of snow or land 704

classifications from MODIS images with the in situ measurements of snow depth ≥ 4 cm. 705

706


37

707

Figure 3. The mean agreements of snow cover classifications for the four snow cover products at 57 708

non-forested and 22 forested stations: (a) snow depth ≥ 4 cm, (b) snow depth =1-3 cm. 709

710


38

711

712

Figure 4. Comparison of mean snow-covered duration (SCD) from HY2000 to HY2010 between 713

MODMYD_MC and in situ observations for all data with snow depth > 0 cm at 57 non-forested 714

stations (a) and 22 forested stations (b). 715

716


39

717

Figure 5. Snow depth and MODMYD_MC retrievals at stations of cultivated farmland (S31), 718

natural grassland (S53) and forest (S12 and S52). MODMYD_MC retrievals are shown as 20 cm 719

for snow and 0 for no snow. 720


40

721

Figure 6. The spatial distribution of mean snow covered days (SCD) derived from MODMYD_MC from HY2001 722

to HY2014 in the Amur River Basin. 723

724


41

725

Figure 7. The histogram of mean snow-covered days (SCD) computed at a 10-day interval from MODMYD_MC 726

during the period from HY2000 to HY2013. 727


42

728

Figure 8. The minimum, mean and maximum monthly snow cover fraction (SCF) during the period from HY2000 729

to HY2013 computed from MODMYD_MC in the entire Amur River Basin. 730

731

732


43

733

Figure 9. SCD anomaly (a), monthly mean SCF (b), SCOD (c) and SCED (d) in the snow-poor year of HY2001. 734

735

a. SCD Anomaly in HY2001 b. SCF in HY2001

c. SCOD in 2001 Fall d. SCED in 2002 Spring


44

736

Figure 10. SCD anomaly (a), monthly mean SCF (b), SCOD (c) and SCED (d) in the snow-neutral year of 737

HY2005. 738

739

740




45

741

Figure 11. SCD anomaly (a), monthly mean SCF (b), SCOD (c) and SCED (d) in the snow-rich year of HY2012. 742

743

744

745

746




46

747 748

Figure 12. The annual snow-covered days (a) and mean snow cover fraction (b) during March and April computed 749

from MODMYD_MC in the Amur River Basin (ARB), Russia and China from HY2000 to HY2013. MA13 refers 750

to March and April of 2013, while being in HY2012. The change rates of SCF are summarized in Table 6. 751

752

753


47

754

Figure 13. Scatter plots of mean Snow Cover Fraction (SCF) and air temperature in March and April from 2001 to 755

2014 in China (a) and Russia (b). 756

757

758


48

759

Figure 14. Annual (a) and Mar-April (b) mean air temperature in China and Russia from 1961 to 2015. 760

761

762


49

763

Figure 15. The mean snow-covered days (SCD), snow cover onset dates (SCOD) and snow cover end date (SCED) 764

in the forested (F) and non-forested (NF) areas in the entire Amur River Basin and in the Sanjiang Plain from 765

HY2000 to HY2013. The change rates are summarized in Tables 7 and 8. 766

767

768


the cryosphere discuss., doi:10.5194/tc-2016-72, …...103 and methodology, presentation of results...

Documents